Permanent Income, Wealth, and Consumption

By Thomas Mayer

This title is part of UC Press's Voices Revived program, which commemorates University of California Press’s mission to seek out and cultivate the brightest minds and give them voice, reach, and impact. Drawing on a backlist dating to 1893, Voices Revived makes high-quality, peer-reviewed scholarship accessible once again using print-on-demand technology. This title was originally published in 1972.

• Language: English
• Publisher: University of California Press
• Release date: Jul 28, 2023
• ISBN: 9780520337169

Thomas Mayer

Enter the Author Bio(s) here.

Book preview

PERMANENT INCOME, WEALTH, AND CONSUMPTION

University of California Press Berkeley and Los Angeles, California University of California Press, Ltd.

London, England Copyright © 1972, by The Regents of the University of California ISBN: 0-520-02103-7 Library of Congress Catalog Card Number: 71-170721 Printed in the United States of America

Contents 1

Contents 1

Preface

Part One

One

TWO

Part Two

Three

Four

Five

Six

Part Three

Seven

Eight

Nine

Part Four

Ten

Eleven

Twelve

Part Five

Thirteen

Fourteen

eFifteen

Sixteen

Appendix One SOME ATTEMPTED TESTS WITH CONSUMERS

Index

Preface

One of the standard issues in empirical economics in recent years has been the testing of the permanent income theory and the life cycle hypothesis. A vast number of empirical tests of these theories have been published. Unfortunately, these empirical tests did not suffice to clarify the issue. In fact, their very large number combined with the disagreements in their results probably means that the whole subject is as much up in the air as it was in the late 1950’s. Even an economist who makes a strenuous effort to keep up with empirical macro-economics is unlikely to have read most of these tests, and of those he has read there are probably many which he has not had time to evaluate adequately. As a result of this there is much confusion in the theory of the consumption function. The plethora of conflicting empirical evidence is such that an economist can decide on the basis of personal likes and dislikes whether he wants to accept or reject these new theories of the consumption function, and he can then find numerous empirical tests which will support and justify his chosen point of view. This is, of course, the opposite of the way tilings should be; empirical tests should limit the opinions which people may justifiably hold.

One of the things I have therefore tried to do in this book is to examine carefully the previous empirical tests of the permanent income theory, the life cycle hypothesis, and related theories. I have been concerned not with describing the theories, but rather with evaluating them, and with distilling from this large body of literature a series of confirmed propositions. This has been a challenge because so many tests reach conflicting conclusions. I have therefore tried to see whether it is possible to explain away this apparent conflict and to show that the valid empirical evidence supports with a unified voice certain propositions. The fact that this has been a feasible task suggests that the current status of empirical economics is not as bad as it seems. Although there is much apparent conflict in the evidence cited by various economists, it has been possible, at least in this one case, to eliminate the conflict.

The key to obtaining agreement between the various tests is to notice that most of those tests which are favorable to the permanent income theory and related theories do not even attempt to show that such a theory is fully valid. Instead, they show that there is a tendency in the direction predicted by this theory. Conversely, those tests which claim to refute the life cycle hypothesis or the permanent income theory do so by showing that these theories are not fully valid. The obvious way to synthesize these tests is therefore by saying that these new theories of the consumption function are partially valid. Such an intermediate position, which for want of a better name I have called the standard income theory, sounds wishy-washy. To make it a serious contender it has to be quantified.

I have therefore undertaken a series of new tests which are concerned primarily with quantifying this standard income theory. These tests, which make up approximately half the book, are based primarily on household budget data rather than on time series data. In undertaking these tests I have focused on tests based on data sources which have not previously been used in this discussion. Lacking the tools of sophisticated econometricians I have generally used very simple techniques. This book is not for anyone whose main interest is in discovering new econometric techniques.

It gives me great pleasure to acknowledge the large debts I have accumulated to many people and organizations in the years I have worked on this book. I am greatly indebted to those who supplied me with unpublished data: Dr. Heiniger of the (Swiss) Bundesamt für Industrie, Gewerbe und Arbeit; M. Perrot of the Institut National de la Statistique et Etudes Economiques; Mrs. Dorothy Projector, Board of Governors, Federal Reserve System; and Mrs. Reddies and Dr. Martin of the (West German) Statistisches Bundesamt, the Institute of Economics and Statistics of Oxford University, and the Survey Research Center of the University of Michigan.

I am also greatly indebted for that necessary ingredient of empirical research, generous financing, to the National Science Foundation, which has shown great patience with continually postponed completion dates; to the American Philosophical Society, whose kind help allowed me to microfilm and process data used in Chapters 13 and 14; and to the Faculty Research Fund of the University of California, Davis. The University of California computer centers at Davis and at Berkeley kindly provided some free time.

Thomas Cargill, Michael Landsberger, Toshiyuki Mizoguchi, and Julian Simon have read much, or all, of the manuscript, and I am deeply indebted to them for their helpful comments. In addition, I received very helpful comments on earlier versions of certain chapters from W. P. Strassmann (Chapters 1, 10, and 16), Andrzej Bizeski, Allan Meltzer, Richard Peterson, Ramachandra Ramanathan, J. Ernest Tanner and Hal Varian (Chapter 13), Margaret Gordon and Albert Fishlow (Chapter 10), and James Holmes, Jacob Mincer, and Margaret Reid (Chapter IS). None of them are, of course, responsible for remaining errors.

I have been greatly aided by a large group of very capable and devoted research assistants and programmers. Over the years they have included James Brundy, Kelvin Calandri, Geoffrey Carliner, Roozbeh Chubak, Noel Luke, Adil Kanaan, John Lee, Hal Varian (who has contributed Appendix II), and Abdolhossain Zahedani. In addition, Alan Cohn, Dennis Fanucchi, Andrea Harzell, Sukhamay Kundu, Lulu Lin, Anthony Mutsear, Richard Piper, Allan Shapiro, David Vanderford, and Gary Walton assiduously calculated seemingly endless columns of figures. Moreover, dealing as I am with much foreign language material, I am indebted to a whole host of translators. I am also grateful to Mr. Gene Tanke for his careful and patient editorial work.

Chapter 10 appeared, in an earlier version, in the American Economic Review, and Chapter 13 in the Journal of Money, Credit, and Banking. I am indebted to both for permission to reprint this material. I would also like to record my debt to Princeton University Press for granting permission to quote Milton Friedman’s A Theory of the Consumption Function numerous times.

Mrs. Mildred Schumway has very ably kept the books on my research grants and related matters. I am indebted to Mrs. Shirley Kelly for typing with great efficiency a manuscript so messy that toward the end, even I could hardly make sense out of it, and to Mrs. Marguerite Crown for retyping seemingly endless revisions.

T. M.

Part One

PRELIMINARIES

INTRODUCTION

This Part consists of two chapters which set the stage for what follows. The first of these chapters discusses the scope of this book as well as some necessary definitions. Chapter 2 provides a brief summary of the theories to be tested. Although some of them, such as Friedman’s permanent income theory and the Modigliam-Brumberg-Ando life cycle hypothesis are well known, others, for example, the Pennsylvania School’s normal income theory, or the Ball-Drake model are less widely known. Since the focus of this book is on empirical testing rather than on theoretical development, I have confined myself to a brief description of the theories.

One

INTRODUCTION

The theory of the consumption function was changed radically in the mid-1950s with the emergence of the new theories of the consumption function—that is, the permanent income theory, the life cycle hypothesis, and related theories.1 Since then there has grown up a substantial body of literature testing and evaluating these theories. But there is still great disagreement about their validity because nobody has taken the trouble to go through these tests systematically, separating valid from invalid tests, and seeing precisely what the valid tests really show.

In this book I shall attempt to synthesize the results of previously published tests to see if one can explain away the disagreement between those test results which claim to confirm the theory and those which claim to disconfirm it I shall also present new tests. The purpose of these new tests is to confirm the results which emerged in previous tests and also to make as specific as I can the synthesis which allows one to reconcile the results of previous tests.

In brief, I will try to show that all the valid evidence on the new theories of the consumption function—the evidence from previously published tests as well as from the new tests presented here—is consistent with a theory which is intermediate between the more traditional approach to the consumption function and the new theories. Such an intermediate approach accepts the proposition that consumption depends on income of several years and that the marginal propensity to consume is greater for permanent than for transitory income. However, it does not support the claims of the new theories that consumption is proportional to permanent income and that it is independent of transitory income. Nor does it accept the very long lag in the consumption function which supporters of the new theories think they have found. Such an intermediate position may sound vague and wishy-washy. But in many of the new tests presented below I am more specific and show that the lag in the consumption function is only about half as long as Friedman claims, and that the marginal propensity to consume permanent income is somewhat closer to the marginal propensity to consume measured income than it is to the permanent income theory’s estimate. Thus while my estimates lie in between those of the new theories and the traditional theories, they lie somewhat closer to those of the traditional theories.2

To establish such results one must examine all the previous tests of the new theories. There is an unfortunate tendency in economics to disregard this step. Someone presenting a new theory simply runs some tests, and if his theory is confirmed by these tests he treats it as generally validated, despite the fact others have previously confirmed rival theories. Such a way of working is only natural. It is usually much more pleasant to work out one’s own tests than to examine carefully other people’s work. And it is natural, too, to give much more credence to one’s own findings than to other people’s. But understandable as it may be, such a procedure is far removed from the standard scientific canon.3 In the physical sciences such an approach is simply not permitted; if an experimenter obtains results which differ from those obtained by others, he will not expect his theory to be accepted unless he can either show an outright fault in previous experiments or show that his experiment is clearly superior to the previous ones. Yet in economics we persist in piling test upon test in the quaint belief that one’s own test— being one’s own—obviously deserves more credence than those tests previously published by others.

The outcome of such an egocentric approach is obvious. Old results do not die, nor do they fade away; they are just buried in an outpouring of new research. The result is confusion. There are a vast number of unevaluated tests in the literature, some supporting the permanent income theory, and some rejecting it. This provides an unfortunate amount of freedom; an economist can decide on very casual grounds whether or not he accepts the permanent income theory, and can then buttress his choice by pointing to a set of tests which support his view, whatever it happens to be. This situation is illustrated by the almost scandalous way in which the profession has treated Friedman’s tests of the perma nent income theory. When Friedman presented his theory, he provided some sixteen tests which confirmed it. Yet economists writing on the permanent income theory have, for the most part, ignored these tests. They have presented new tests and have assumed that if these new tests disconfirm the permanent income theory then the theory is invalidated—even though Friedman’s many tests, which they have no reason to reject, confirm it.

At first glance such a procedure may seem justified. It is an old bromide that it takes just one unfavorable piece of evidence (fact) to disconfirm a theory regardless of how well the theory has performed on other tests. This is an inadequate description of how scientists actually work,4 and furthermore it is not applicable in a situation where there are only two possibilities—for example, either consumption is proportional to permanent income or it is not. For instance, one of the pieces of evidence used by Friedman is his demonstration that if the permanent income elasticity of consumption were actually less than unity, then we would find a tendency for the distribution of income to become more unequal over time. And since we do not observe such a tendency, the income elasticity of consumption cannot be less than unity.⁰ Suppose now that someone undertakes a new test which shows that the income elasticity of consumption is less than unity. This new test alone does not permit us to reject the proposition that the permanent income elasticity of consumption is equal to unity, for the rival proposition that it is less than unity is disconfirmed by Friedman’s test. Gearly, this conflict of evidence can be resolved only by critically examining both pieces of evidence.

Moreover, there is another point. The new theories of the consumption function, as well as the traditional theories, are paradigms for which their supporters have provided much empirical evidence. Thus, when a supporter of one of these paradigms is confronted with a new test which claims to contradict his paradigm, he is often ready to reject the new test without much ado because he has the evidence of so many other tests to support his view. Hence, new tests of the permanent income theory or of the life cycle hypothesis are not likely to change many opinions among those already committed. The way to settle the dispute is not merely to undertake new tests, but also to examine critically the evidence previously provided by the supporters and the critics of the new theories.

I have therefore devoted about half of this book to examining previous tests. In doing so, I have been concerned with evaluating these tests rather than with describing them. Since most of them can be found in readily available sources, there is little purpose in describing them in detail. I tried to provide just enough of a description so that the reader can follow my evaluation.

Such an evaluation of the previous literature must be comprehensive. To examine only a few tests is not sufficient because any test which is omitted may be just the one which provides strong confirmation or disconfirmation of the permanent income theory or a related theory. I therefore intended to cover all the existing tests, but I have not quite succeeded in this. For one thing, I have not been able to locate copies of some unpublished tests. In addition, I had to pass up some Japanese tests because of the language problem. Third, I have not dealt with consumption functions fitted to foreign time series except those which had special relevance for the wealth theories or had been fitted to a cross-section of countries. My lack of familiarity with foreign data would have made any attempt to evaluate all foreign functions very time-consuming. Finally, I am sure that I have inadvertently overlooked some tests. Hence, if the reader notices that I have omitted a certain test, he should interpret this not as an unfavorable reflection on the validity of this test, but rather as a reflection on my own research.⁸

Even so, there are many tests to be covered—many more than I thought there were when I started on this book. Some readers may find this long journey through so many tests tedious. In part for this reason, I have provided at the end of Chapters 3-6 summary tables which list the tests, classify them as valid or invalid, and give a brief indication of what they show. Any reader wishing to skip some of the discussion of these tests, and willing to take my judgment on faith, can simply turn to these tables.

After evaluating the existing tests I present my own tests. I have tried to undertake as varied a set of tests as I could. I very much doubt that it is possible to resolve the debate by any single crucial experiment. Definitive tests are rare—perhaps even nonexistent—in economics, and it is highly unlikely that as complex an issue as the validity of the permanent income theory could be settled by any single test. Nearly every test one can think of suffers from some weakness, and even if the test procedure is above criticism, one can always criticize—and usually with considerable justification—the underlying data.⁷ Hence, the best way of

6. Two as yet unpublished manuscripts reached me too late for inclusion. They are A. S. Deaton, Wealth Effects on Consumption in a Modified Life- Cycle Model, and W. H. Somermeyer and K. van de Rotte, A Macro- savings Function for the Netherlands, 1949-1966.

7. For example, when confronted with a test using Israeli data which contradict the proportionality hypothesis, Friedman did not abandon this proceeding is to undertake many tests. If one runs a number of tests and finds that they all agree, then one can accept the result of these tests even if each of them suffers from some different defect To set this forth more formally, consider a test of the proposition that A > B. Suppose that this test does suffer from some defect, but that this defect is not very great. If this test shows that in fact A > B then there are two possibilities; either this proposition is true or else the result obtained is due to the defect in the test. Suppose that before looking at this test one assigns an a priori probability of, say 80 percent to A < B, and that one assigns an a priori probability of 25 percent to the proposition that the defect of the test is large enough to lead to the wrong result In this case, when confronted with such a test which shows A > B, it is rational to reject the test and to continue to believe that A < B. But suppose that there are five such tests and that they each suffer from quite different defects. If all of these five tests, though admittedly defective, agree in showing that A > B, then it is rational to reject die hypothesis that A < B even though none of the tests individually would make one do so. This is really nothing new; all it means is that circumstantial evidence can be convincing if there is enough of it It may be useful at this point to outline what is to follow. Chapter 2 provides a summary of the various theories to be tested. Chapter 3 starts the evaluation of previous tests with a detailed discussion of the various tests presented by Milton Friedman in his A Theory of the Consumption Function. The following three chapters deal with the tests of the permanent income theory undertaken by others. Part III (Chapters 7-9) then deals with numerous consumption functions which have been fitted to aggregate time series. Chapters 7 and 8 subject many of these consumption functions to a simple naive model test. Chapter 9 then takes the survivors of the naïve model test and examines how they rank when used in a projection test. In addition, Chapter 9 gives the results of using different methods of regressing consumption on income. Part IV (Chapters 10-12) presents a series of tests using as permanent income proxies, occupation, race, and location. The final part then consists of three chapters using budget study data in other tests plus a summary chapter. Chapter 13 deals with a small sample of households for whom income and consumption data are available for five years. Chapter 14 uses a handful of budgets which cover a longer time span.

hypothesis but suggested that he would do so only if this test would yield the same result when applied to other data. (Note on Nissan Liviatan’s Paper, in Carl Christ (ed.), Measurement in Economics; Studies in Mathematical Economics and Econometrics in Honor of Yehuda Grunfeld (Stanford, Stanford University Press, 1963) p. 63. Similarly, in my discussion below I am rejecting one test, a test which does not fit any known theory of the consumption function, on the grounds that the data are bad.

Chapter IS then uses data on the net worth of households to test the proportionality hypothesis. The final chapter provides a summary of the results. Appendix I then describes an unsuccessful test. But before proceeding with this material, problems of method and definition must be discussed.

One of the most radical claims of the permanent income theory and of the life cycle hypothesis is that consumption is proportional to permanent income—that is, that the permanent income elasticity of consumption is equal to unity. Although this is the most fascinating hypothesis of the new theories—and one which I have tried to test in numerous ways—it is by no means clear that this hypothesis is really testable with presently available techniques. The only way to test this hypothesis is to compare households with different permanent incomes either at one moment of time (cross-section tests) or over time (time series tests). Such comparisons assume that one can isolate the effects of differences in permanent income. But many other factors, apart from permanent income differences, can affect consumption; and if some of these factors are correlated with permanent income then there is the danger of obtaining a spurious correlation between permanent income and the propensity to consume.

This type of problem is, of course, a frequent one in economics and the way to guard against it is to include such factors as separate variables in the regression, or at least to look at them and see the direction of the bias they introduce. Unfortunately, the necessary data are frequently not available. The variable which is probably the most troublesome in this way is the tastes factor. If differences in the taste for savings are correlated with permanent income, then the usual type of regression we run in economics may attribute to differences in permanent income what is really the effect of differences in tastes.

This is a serious problem because there is some sociological evidence that different income classes do have different tastes for saving. As Pierre Martineau has remarked:

It seems that many economists overlook the possibility of any psychological differences between individuals resulting from different class memberships. It is assumed that a rich man is simply a poor man with more money and that, given the same income, die poor man would behave exactly like the rich man. The Chicago Tribune studies crystallized a wealth of evidence from other sources that this is just not so, and that the Lower-Status person is profoundly different in his mode of thinking and his way of handling the world from the Middle-Class individual. … Lower-Status people typically explain why one should save—why the act of saving is important. On the other hand, Middle- Class people do not, as if saving is an end-in-itself, the merits of which are obvious and need not be justified. … Here are some psychological contrasts between the two different social groups: Middle Class: pointed to the future. His viewpoint embraces a longer expanse of time. … stresses rationality. … Lower status: pointed to die present and past, lives and thinks in short expanses of time. … nonrational essentially. … very much concerned with security and insecurity.⁸

Since there exists at present no adequate way of measuring tastes, there are only three possible ways of handling this problem. One is to abandon all hope, to conclude that the proportionality hypothesis, at least at present, is not refutable. And since (as will be shown below) there is no valid evidence supporting it either, there is nothing one can say about it. Thus, if one takes this approach, one of the important issues in consumption function theory would have to be shelved until one has a good way of measuring tastes.

The second alternative is to argue that while the taste variable may have some effect, its effect is so small that it can be ignored. While this approach is certainly convenient, there is no evidence that it is correct.

But there is still another approach. This is to say that if differences in permanent income are always found in combination with differences in tastes, then let us treat permanent income and tastes as one single variable. This procedure is, of course, most defensible if differences in tastes are themselves caused by changes in permanent income, and if the relationship between permanent incomes and tastes is a stable one.

Fortunately, the fact that one is combining an income and a tastes variable does not reduce the applicability of the new theories to some practical problems; in fact, the opposite is sometimes the case. Thus, consider one of the applications of the permanent income theory discussed by Friedman: evaluating the effect of income redistribution as a way of changing the average propensity to consume.⁸ Clearly, what is relevant here is the existence of differences in the average propensity to consume of various permanent income classes, regardless of whether these differences are due to differences in permanent income per se, or to differences in tastes.

The same may be true for long-run predictions of the saving-income ratio. If the correlation of savings tastes with permanent income is the same for secular changes in permanent income as it is for cross-section differences in permanent income, then no harm is done, if, in the cross- section data, one has estimated the combined effects of permanent in-

8. Social Classes and Spending Behavior, reprinted in Martin Grossack (ed.), Understanding Consumer Behavior (Boston, Christopher Publishing House, 1964), pp. 134—47. See also Louis Schneider and Sverre Lysgaard, The Deferred Gratification Pattern, American Sociological Review, Vol.

18, April 1953, pp. 142-49, and the literature cited therein.

9. Milton Friedman, A Theory of the Consumption Function (Princeton, Princeton University Press, 1957), p. 237.

come and tastes. To be sure, the correlation of permanent incomes and tastes may well be different for secular changes in permanent income than it is in cross-section studies. But there is little reason to think that tastes do not change at all with secular changes in permanent incomes. Hence, even if one could somehow hold tastes constant and estimate the pure relationship between permanent income and the average propensity to consume, this relationship may well be a worse guide to the effects of secular income changes than the combined effect of changes in permanent income and in tastes.

I have therefore used cross-section tests, which necessarily combine taste effects with income effects, in my tests of the proportionality hypothesis. This may seem improper since the proportionality hypothesis is formulated in the new theories without explicit reference to tastes. But the adherents of the new theories have implicitly included tastes along with permanent income effects in their own empirical tests and, what is more, they generally do not warn the reader of any potential bias created by the taste effect. Hence, they implicitly combine the direct effect of permanent income, and its indirect effect through taste differences, into one single variable they call the effect of differences in permanent income.5 Moreover, as already pointed out, when discussing the policy implications of his theory, Friedman did not separate taste effects from income effects either.

So much for the proportionality hypothesis. Another problem of testability arises in connection with Friedman’s hypothesis that the marginal propensity to consume transitory income is zero. But since this problem can be discussed best in direct connection with Friedman’s theory, I have deferred discussion of it until the next chapter.

Another problem relates to the definition of consumption. The new theories of the consumption function generally define consumption in the use sense; that is, they count as consumption not the purchase of consumer goods, but only their actual destruction by use. (In principle this treatment should be extended to all goods; in practice, due to data problems, the adherents of the new theories confine it only to consumer durables.) Therefore, they count as consumption only the purchase of nondurables plus the depreciation on the household’s stock of durables. In this book I have not done this, but rather have defined consumption in the conventional way as the purchase of consumer goods. To evaluate a theory which defines consumption in one way by tests which define it in another way may seem silly, but there is a good reason for it: the scarcity of data for the use concept of consumption. This fact shows up not only here, but also in the work of the adherents of the new theories. They, too, used many tests based on the conventional definition of consumption. In fact, this is the case for the great majority of the tests used by Friedman himself. And, as Friedman has remarked: I cannot very well accept [a test using the conventional definition of consumption] when it works and reject it when it does not. ¹¹ The adherents of the new theories have essentially used both definitions of consumption, one in the formal exposition of their theories, and the other in much of their testing of these theories. Since I am concerned here with the empirical tests of the theories, my use of the conventional definition of consumption does not really depart very far from practice of the supporters of the new theories.

As mentioned above, my reason for choosing the conventional definition of consumption is mainly that this is the definition for which many more data are available. Household budget surveys, the source of most of my data, do not employ the use definition of consumption, and had I decided to work with this definition I could only have employed tests using aggregate time series data.¹³ As will be shown in Chapter 9, such tests are hardly conclusive.

11. Savings and the Balance Sheet, Oxford University Institute of Economics and Statistics, Bulletin, Vol. 19, May 1957, p. 126.

12. Some economists have tried to avoid his problem by using nondurable expenditures. But defining consumption as nondurable expenditures does not provide a good proxy for the use definition of consumption since it excludes the depreciation of the household’s stock of durables. As a rough approximation, one can assume that the household’s stock of durables is constant—that is, that depreciation is equal to new purchases. If this is even approximately so, then the conventional definition of consumption provides a better estimate of the use definition of consumption than does the definition of consumption as nondurable consumption. For any test of the proportionality hypothesis the conventional definition is, therefore, probably substantially superior to the nondurables definition of consumption. However, for testing the hypothesis that consumption out of transitory income is zero, the nondurables definition may be better. This is so because depreciation of a household’s stock of durables may bear little relation to its transitory income, so that the marginal propensity to consume transitory income through depreciation may be close to zero. Hence for the marginal propensity to consume transitory income, the use definition of consumption is perhaps approximated better by the nondurables definition than by the conventional definition. But the superiority of the nondurables definition on this score is not very important, for as will be shown below, the hypothesis that the marginal propensity to consume transitory income is zero is not really testable. Turning to the third hypothesis of the new theories, the lag hypothesis, the nondurables definitions, like the conventional definition, is bad because nondurable consumpion may have a different lag than use consumption. (It is not true, incidentally, that nondurable consumption necessarily has a shorter lag than total use consumption. One of the important items of nondurable consumption, rent, presumably has a long adjustment lag.)

In any case, I believe that little is lost by using the conventional definition of consumption. The permanent income theory claims that when consumption is defined in the use sense, then the permanent income elasticity of consumption is unity and the transitory income elasticity of consumption is zero. A definition which makes important parameters equal to unity and zero is convenient, and if the permanent income theory were actually correct, then the use definition of consumption might be the better one. But, as will be shown below, even for the use definition of consumption, there is no evidence that it is indeed valid.

Quite apart from the new consumption function theories, the use definition of consumption is frequently considered to be superior to the conventional definition because it corresponds more precisely to the meaning attached to consumption in economic theory.¹⁸ This idea is certainly true as far as it goes, but it neglects the fact that the use definition is still very far from the theoretically correct one. One reason for this is that to apply the use definition of consumption correctly, it is not sufficient to subtract from total consumption expenditures on consumer durables as defined in the data. One would have to subtract certain semi-durables and even some nondurable goods and services.¹⁴ The data necessary to do this are generally not available. Second, the use definition of consumption ignores the fact that certain consumer expenditures, such as some expenditures on education, should not be included in consumption. Third, intergenerational transfers create a problem, one which is almost unmanageable on an empirical level. Insofar as families expect their children to care for them later in life, some of their expenditures on children should be treated as saving rather than as consumption.¹⁵ And similarly some of the expenditures which children undertake on behalf of their parents should be considered as repayments of debts, that is, as saving.¹⁹ Thus, even if one measures consumption in accordance with the use definition, the

13. Moreover, households may think of their durable purchases not as consumption but as saving.

14. On services as durables, see Milton Friedman, The Optimum Quantity of Money (Chicago, Aldine Publishing Co., 1969), p. 49.

15. Thus Friedman has argued that expenditure on children is part of saving. Reply to Comments on ’A Theory of the Consumption Function,’ in Lincoln Clark (ed.), Consumer Behavior (New York, Harper and Brothers, 1958), Vol. 3, p. 463.

16. Moreover, there is the fact that our household data do not include capital gains and gifts in income. Hence, if consumption is derived by subtracting saving from income, as is frequently the case, for some households consumption is seriously overstated. (And since households receiving capital gains tend to be high-income households, this overstatement may create a serious bias.) resulting data are far removed from what is called consumption in economic theory. Employment of the use definition in place of the conventional definition moves one only part of the way toward the correct definition.

To be sure, half a loaf is better than none, and if the use definition did not suffer from any disadvantages, it would, despite the above shortcomings, be superior to the conventional definition. But the use definition does in fact have serious shortcomings. One obvious one, already discussed, is the absence of household data. But even if the data were available, one might still prefer the conventional definition of consumption. The conventional definition relates consumption to actual expenditures, and so it relates directly to the demand for labor. This makes it useful for many practical macroeconomic problems. Thus, when Friedman came to discuss the practical implications of his theory, these turned out to be situations in which the conventional definition of saving is generally more relevant than the use definition.¹⁷

On the other hand, it must be admitted there is something to be said for the use definition, too. If one is concerned with measuring consumption from a welfare point of view then the use definition is superior to the conventional definition.¹⁸

Finally, there is another point. This is that the adherents of the new consumption function theories, even when they have employed the use definition of consumption in their empirical work, have not been really consistent. If one includes the depreciation of durables in consumption, then one should also include the services which the household obtains from its stock of durables in income. None of them have done this. Hence, when they have regressed use consumption on conventionally measured income, they have employed neither the use definition nor the conventional definition in a logically consistent manner, but have really worked with a peculiar hybrid which lacks any theoretical significance.

I hope that this has convinced the reader that it is legitimate to follow Friedman’s usual procedure and to test the new consumption function theories by data employing the conventional definition. If not, the reader can look upon the tests presented below as tests not of the permanent income theory and life cycle hypothesis per se, but as tests of a series of propositions which differ from these theories in their definition

17. Milton Friedman, A Theory of the Consumption Function, op. cit., pp. 233-39.

18. Another issue is raised by the practical problem of predicting consumption expenditures. It may well be the case that one can predict consumption better by predicting durable and nondurable expenditures separately. But this procedure does not correspond to the use definition of consumption since it leaves out depreciation on durables.

of consumption. Since these propositions are important ones—regardless of whether or not one treats them as parts of the new consumption function theories—they are worth testing in any case. Moreover, insofar as my results are invalid as tests of the new consumption function theories, this applies only to the new tests I present below. In my evaluation of the previous tests, I am using the same consumption concept used originally in these tests. Hence, one of my conclusions—that there is no evidence at all supporting the full permanent income theory and life cycle hypothesis—is unaffected by this definitional problem.

Although there is much to be said against coining new terms, it is worthwhile to develop a few special terms to save verbiage in what follows. I will use the phrase, measured income theories to denote the absolute income theory and the relative income theory.¹* I will use the term wealth theories not only as a generic term to refer to Friedman’s theory but also as a broader term to include the life cycle hypothesis and the rest.⁹⁰ The term strict or full permanent income theory will designate Friedman’s theory in distinction to the standard income theory or the life cycle hypothesis. Finally, unless otherwise specified, when referring to wealth, I will mean nonhuman wealth only.

19. Above I have referred to these theories as traditional theories. The term measured income theories is better since the term traditional theories could also be taken to mean the theory developed by Irving Fisher, a theory which in many ways is the same as the permanent income theory. I have used the designation measured income theories because in contrast to the permanent income approach, they assume that consumption is a function of measured income.

TWO

THE THEORIES

Before taking up the various tests of the permanent income approach, it is necessary to be clear about what is being tested. This chapter gives a brief description of the various theories encompassed by the term wealth theories. Since I am concerned in this book with empirical tests I will describe these theories only to the extent needed to understand the empirical tests and will not take up the refinements which have appeared in the literature in recent years, often in connection with growth models. But to see these theories in the proper context, one must first look at their background—that is, at the theories prevailing at the time when the permanent income approach was developed.

Previous Theories

As is so frequently the case, a discussion of a modern economic theory should start with the work of Irving Fisher. Fisher described the factors causing households to save in the context of a multi-period model.6 Two characteristics of Fisher’s model relate it to the modern wealth theories. One is Fisher’s explicit focus on the rationality of savings decisions. Fisher’s households save purposefully. This contrasts with the subsequent treatment of saving by Keynes. The second way in which Fisher’s theory resembles the contemporary wealth theories is that Fisher explicitly used a multi-period model and showed no sympathy for the notion that saving can be explained by current income alone.

Although Fisher’s work is sometimes looked upon as the fountainhead of the permanent income theory, what Fisher taught was not the strict permanent income theory. He strongly rejected one of its important components, the proportionality hypothesis:

In general, it may be said that, other things being equal, the smaller the income, the higher the preference for present over future income. … It is true, of course, that a permanently small income implies a keen appreciation of future wants as well as of immediate wants. … This influence of poverty is partly rational, because of the importance, by supplying present needs of keeping up the continuity of Ufe and thus maintaining the ability to cope with the future; and partly irrational, because the pressure of present needs blinds a person to the future.²

Fisher’s discussion of saving was more or less brushed aside by the Keynesian revolution. To be sure, in the General Theory Keynes did devote three pages to a discussion of savings motives, but the main thrust of Keynesian theory dealt with consumption rather than with saving. Hence, there is a tendency in the Keynesian literature to treat saving as a residual rather than as something to be explained by a multi-period model. And if one focuses on consumption rather than on saving then one is much less likely to concern oneself with the problem of the relevant time period than if one focuses on saving. This is so because although the motives for saving necessarily involve more than one period, the motives for consumption relate to the current period. By stressing consumption rather than saving, Keynes set the stage for ignoring the question of the time period. Hence, economists tended to follow the line of least resistence by using the unit of time in which the data happen to come—usually a year.

Although Keynes did not stress the savings motives discussed by Fisher, he did follow Fisher in rejecting the proportionality hypothesis:

The fundamental psychological law upon which we are entitled to depend with great confidence both a priori from our knowledge of human nature and from the detailed facts of experience, is that men are disposed, as a rule and on the average, to increase their consumption as their income increases, but not by as much as the increase in their income. … This is especially the case when we have short periods in view, as in the case of so-called cyclical fluctuations of employment during which habits, as distinct from more permanent psychological propensities are not given time enough to adapt themselves to changed objective circumstances. For a man’s habitual standard of life usually has the first claim on his income, and he is apt to save the difference which discovers itself between his actual income and the expense of his habitual standard.³

While Keynes’ statement that past consumption habits affect current consumption may, at first glance, suggest something fairly similar to the permanent income theory, it is really a quite different theory, a habit persistence theory. It is previous consumption rather than long

run income which Keynes had in mind. He was most explicit on this point.7

But while the absolute income theory, as Keynes’ formulation came to be called, differs fundamentally both from die preceding theory of Fisher and from the subsequent permanent income theory, Keynes did employ certain elements of the wealth theories. One is that he considered changes in the stock of wealth to have an important effect on consumption.8 Another is that he thought the propensity to consume to be much less for transitory income (windfalls) than for permanent income.⁴ Although these qualifications blur the sharp edges of the contrast between the absolute income theory and the permanent income approach, there does remain the fact that Keynes brushed aside, without much ado, the logically compelling analysis of Fisher. Excessive regard for the writings of other economists is hardly a characteristic of the General Theory.

In the 1940s the credibility of the absolute income theory was severely challenged by two developments; one was the failure of the theory’s forecasts of postwar consumption, and the other was Kuznets’ demonstration that the average propensity to consume had not risen secularly. These developments set the stage for the emergence of the relative income theory, which reached its full formulation in the work of Franco Modigliani and James Duesenberry.9

The relative income theory treats the proportionality hypothesis in a quite different way from Fisher and Keynes. It does not assume that a low income, in and of itself, makes households less likely to save. If all households lived in a vacuum, then their average propensities to consume would be independent